I wonder how to reproduce the result in pre_dav_ytb. I directly use the initial setting to train(stage=pre_dav_ytb, batch_size=32,iteration=1e5) but the testing result is always about 2% below the result in the paper. Also, I find that in the paper it is said that step of davis is 50000 while the step of youtube is 1e5, but in the code the step of pre_dav_ytb(Which in my understanding is the combination of davis training and youtube training) is only 1e5 (I am expecting 1.5e5). So I wonder what is the correct way to reproduce the result?
I wonder how to reproduce the result in pre_dav_ytb. I directly use the initial setting to train(stage=pre_dav_ytb, batch_size=32,iteration=1e5) but the testing result is always about 2% below the result in the paper. Also, I find that in the paper it is said that step of davis is 50000 while the step of youtube is 1e5, but in the code the step of pre_dav_ytb(Which in my understanding is the combination of davis training and youtube training) is only 1e5 (I am expecting 1.5e5). So I wonder what is the correct way to reproduce the result?